A settlement meant to compensate authors has turned into an argument about who qualifies as one. In the same stretch of weeks that writers started publicly questioning why their publishers were claiming a cut of Anthropic settlement money on decades-old books, scammers began flooding author inboxes with fraudulent emails impersonating the United States Copyright Office. Money moves, and the ecosystem around it wakes up.
I review AI tools for a living, which means I spend most of my time looking at what these systems can do. This story is about something else: what happens downstream when a company that trained on copyrighted books has to write checks. And the answer, so far, is messier than anyone selling you an AI ethics narrative wants to admit.
The dispute nobody planned for
Authors have been posting on social media that publishers appear to be claiming more than their fair share of certain payments. Mystery and thriller writer April Henry raised the issue publicly about HarperCollins. The pattern authors describe involves older works, the kind of backlist titles where rights arrangements were signed long before anyone imagined a language model ingesting the text.
That timing detail matters more than the outrage. Contracts written in 1998 or 2007 did not have a clause covering “compensation received from an AI developer for unauthorized training use.” So when a settlement fund arrives, both sides can read the same contract and reach opposite conclusions in good faith. Publishers see a subsidiary rights payment tied to a work they invested in. Authors see damages for their own words being taken, with a publisher inserting itself into a harm it did not suffer.
Agents sit in an equally awkward spot, since commission structures were also built for a world of advances, royalties, and foreign rights deals rather than class action distributions.
Why this was predictable
Every conversation about AI and copyright over the past few years has focused on the wrong question. The debate ran on whether training on books without permission is legal, as if establishing liability would automatically produce justice for writers. It does not. Liability produces a pile of money, and a pile of money requires a distribution mechanism, and distribution mechanisms get captured by whoever has the best lawyers and the clearest paperwork.
Individual authors have neither. They have contracts they signed under time pressure, often without legal review, frequently with clauses they did not fully parse. Publishers have rights departments. That asymmetry did not appear because of AI. AI just gave it a fresh occasion to express itself.
The uncomfortable version of this story is that authors won a legal argument and may still lose the economics of it. Not because anyone committed fraud, but because the money is flowing through pipes that were never designed to carry it.
The scammers arrive on schedule
The fraudulent Copyright Office emails deserve their own paragraph, because they tell you something about how this space actually works. Fake messages asking authors to verify copyright registrations are landing right as writers are genuinely uncertain about their rights status, genuinely watching for official-looking correspondence, and genuinely hoping money is coming.
That is not coincidence. Direct solicitation has been a primary recruiting method for publishing scams since the mid-2010s. Confusion is the raw material. Any moment when authors do not know what legitimate contact looks like is a moment scammers can occupy. A settlement process generating unfamiliar paperwork is close to ideal conditions.
Practical guidance, since this is the part that might actually save someone money:
- The Copyright Office does not email you asking to verify a registration through a link. Treat any such message as fraudulent by default.
- Verify settlement communications through the official claims administrator, not through whatever arrives in your inbox.
- If your publisher or agent claims a share of a payment, ask them to point to the specific contract clause they are relying on. In writing.
- Talk to other authors on the same imprint. Patterns are easier to see across multiple contracts than inside one.
What I take from this
My beat is telling you whether a tool is worth your money and your trust. This story is a useful reminder that the trust question extends past the product. A model trained on unlicensed books eventually generates a legal bill, and that bill gets settled through institutions with their own incentives. The writers whose work was used are the last party in line and the least equipped to argue.
Anyone promising that AI companies paying up will fix things for creators is selling a simpler story than reality supports. Compensation is not the same as fairness, and a settlement fund is not the same as a functioning market for training rights. The first is a one-time cleanup. The second would require contracts that name AI use explicitly, and nobody has been in a hurry to write those.
Until then, expect more of this: authors discovering that winning creates a new negotiation, and opportunists showing up for the confusion in between.
🕒 Published: